A Genetic Programming Approach to Designing CNN Architectures, In GECCO 2017 (oral presentation, Best Paper Award)
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
.gitignore
LICENSE
README.md
cgp.py
cgp_config.py
cnn_model.py
cnn_train.py
exp_main.py

README.md

Designing Convolutional Neural Network Architectures Based on Cartegian Genetic Programming

This repository contains the code for the following paper:

Masanori Suganuma, Shinichi Shirakawa, and Tomoharu Nagao, "A Genetic Programming Approach to Designing Convolutional Neural Network Architectures," Proceedings of the Genetic and Evolutionary Computation Conference (GECCO '17), pp. 497-504 (2017) [DOI] [arXiv]

Requirement

We use the Chainer framework for neural networks and tested on the following environment:

  • Chainer version 1.16.0
  • GPU: GTX 1080 or 1070
  • Python version 3.5.2 (anaconda3-4.1.1)

PyTorch version is here

Usage

Run the architecture search

This code can reproduce the experiment for CIFAR-10 dataset with the same setting of the GECCO 2017 paper (by default scenario). The (training) data are split into the training and validation data. The validation data are used for assigning the fitness to the generated architectures. We use the maximum validation accuracy in the last 10 epochs as the fitness value.

If you run with the ResSet described in the paper as the function set:

python exp_main.py -f ResSet

Or if you run with the ConvSet described in the paper:

python exp_main.py -f ConvSet

When you use the multiple GPUs, please specify the -g option:

python exp_main.py -f ConvSet -g 2

After the execution, the files, network_info.pickle and log_cgp.txt will be generated. The file network_info.pickle contains the information for Cartegian genetic programming (CGP) and log_cgp.txt contains the log of the optimization and discovered CNN architecture's genotype lists.

Some parameters (e.g., # rows and columns of CGP, and # epochs) can easily change by modifying the arguments in the script exp_main.py.

Re-training

The discovered architecture is re-trained by the different training scheme (500 epoch training with momentum SGD) to polish up the network parameters. All training data are used for re-training, and the accuracy for the test data set is reported.

python exp_main.py -m retrain